Intel Axes Rialto Bridge GPUs, Delays Falcon Shores to 2025
The VP and GM of Intel’s Supercomputing Group announced today via a blog post that Intel is canceling its upcoming Rialto Bridge series of data center GPUs and moving data center GPU releases to a two-year cycle. Did. The company’s next data center GPU offerings will therefore come in the form of hybrid chips based on the Falcon Shores chiplet, though the blog states that these are expected to go into production in 2025. This puts him a year behind Intel’s previous predictions for 2024. Shores XPU can combine both CPU and GPU technology in his one mix-and-match chip package, but is expected to arrive first as a GPU architecture in 2025.
Additionally, Intel is canceling Lancaster Sound GPUs slated for its Flex series of data center GPUs. Instead, Intel will focus on the next-generation Melville Sound products in its series.
According to Intel, the new cadence of releases is based on customer expectations for their datacenter GPU offerings and is largely in line with the incremental launch rates seen from other GPU companies such as Nvidia. The move follows Intel’s recent restructuring of his AXG graphics group to address the gaming and data center markets separately by placing them under two separate business units of his. everything.
Intel also said that its Xeon and process node roadmaps are both on schedule. Intel has said it will improve its focus on its software ecosystem and offer continuous updates to its Max and Flex series GPUs that include more performance, features, and expanded operating system support.
Intel’s Falcon Shores XPUs are the company’s ultimate goal of delivering 5x the performance-per-watt of existing server chips, 5x the compute density on x86 sockets, and 5x the memory capacity and bandwidth. represents a continuation of the heterogeneous architectural design arc of Intel’s roadmap for CPUs and GPUs for HPC is converging on Falcon Shores, indicating that these chips will do both in the future.
This fragmented chip design has separate tiles for x86 compute cores and GPU cores, but Intel uses these tiles to match all CPU models, all GPU models, or a mix of the two. You can create any mixture of the two additional elements, such as Intel says these tiles will be manufactured on an unspecified Angstrom-era process node, but Intel’s 20A seems to fit the bill for tiles they can manufacture themselves.
Intel’s current-generation Ponte Vecchio GPU was to be followed by the Rialto Bridge, its next-generation data center GPU due in 2023. Intel has revealed that the chip will feature up to 160 Xe cores. Ponte Vecchio. The chip is said to include unspecified architectural enhancements similar to a “tick” that will improve application performance by up to 30% over the Ponte Vecchio.
The chip also benefited from a higher power envelope — Intel lists the Rialto Bridge’s peak power consumption at 800W, which is above the Ponte Vecchio’s 600W peak. Intel has said it will adopt the OAM 2.0 specification, but will continue to offer GPUs in other form factors as well. The Rialto Bridge was compatible with the Ponte Vecchio package, so it could be a drop-in upgrade.
This is breaking news… more to come.